Introduction - It’s the remix!

The corpus I chose for my Computational Musicology portfolio is a selection of remixes of (pop) songs that I like. I’ve been maintaining a Spotify playlist, that now has around 75 remixed recordings, since last year. I started maintaining this playlist for myself because there are generally a lot of remixes for artists that I follow, but I was unfamiliar with most. They’re quite an underappreciated part of an artist’s discography. Listening to a lot of remixes and collecting those that I like was an interesting musical journey.

It’d be interesting to find out if there’s some general elements shared among the remixed recordings that I like. Comparing the remixes to the original recordings can also be interesting, for sure the tempo would probably be higher for the remixes. The definition of a remix is sometimes confusing, but the ones on my playlist are either extended remixes by the original artists or a remix where a significant amount of production on the song has been altered from the original. Remixes where the only difference is a new guest artists are not included on the playlist.

The corpus is a personal playlist, artists on the playlist include Lady Gaga, Dua Lipa and Charli XCX, among others. Since the corpus is my personal playlist, it doesn’t fully cover the pop remixes genre, and there aren’t that many remixes of top 10 hits included.

Typical tracks:

  1. Katy Perry - Chained To The Rhythm (Oliver Heldens Remix)
    • This is quite a typical pop remix, as there are lots on Spotify. Producers like Oliver Heldens, R3HAB and more just love to put these out. This is the type of remix that does no harm, but also doesn’t bring a lot (different) to the table.
  2. Lady Gaga & Ariana Grande - Rain On Me (Purple Disco Machine Remix)
    • This is basically an extended version of the original, but the beat is a bit more house. There are more extended versions/remixes of tracks on this playlist, so it’s quite typical.

Atypical tracks:

  1. Lady Gaga - Stupid Love (BloodPop® & BURNS Vitaclub Warehouse Remix)
    • This is almost an entirely different song from the original. It’s a lot more techno-adjacent and the structure is completely different.
  2. Kesha - Praying (Frank Walker Remix)
    • This turns an acclaimed ballad into a dancefloor banger, with a good drop and build-ups. It may sound typical, but it feels so different and flips across genres.

How remixes switch up the tempo.


Tracks are remixed for a large amount of reasons, an important one being: making a song more suitable for the dancefloor. This means there is probably some change in tempo.

In the violin plot, which shows the full tempo distribution for both categories, it can be seen that for original recordings, the tempo for most tracks lies between 100-140 bpm, with a peak around 120 bpm. The tempo distribution for their remixed counterparts is far more dense, with a peak just slightly above 120 bpm. Most house music, and thus, remixes that fall into that genre, is around 128 bpm nowadays, which can be significant factor in the overall tempo change.

Tempo alone doesn’t define how club-ready a track is, two other important factors are energy and danceability.

Energy and danceability; are “my” remixes more club-ready?


This graphic shows the Spotify API-determined energy and danceability for all tracks in the corpus. Most original recordings in the corpus were already quite high in both energy and danceability, with almost all tracks being in the upper right quadrant.

Overall, energy and danceability are somewhat higher among the remixes. Their energy is distributed within the same range as the original, but a shift to the right is apparant. The danceability is not distributed within the same range, there are a more tracks with less than 0.5 danceability, most of those having higher energy.

When hovering over a point in the graphic, the exact values and song title are shown. The track’s counterpart in the other category gets highlighted as well, highlights can be made undone by doubleclicking on whitespace in the graphic.

Different mix and different energy.


Energy is an important factor, and listening to the remixes with a lot of change in the assigned energy value confirmed that. Seems that in most cases, remix with a (big) change in energy are actually very different songs.

So, what about the remixes with the smallest change in energy?

Top 10 remixes with least change in energy, with duration change
Remix Energy Change in energy from original Change in duration (seconds)
Walking Away - Mura Masa Remix 0.528 -0.003 51.299
Circus - Villains Remix 0.739 0.006 125.200
Simmer - Caroline Polachek Remix 0.606 0.007 -69.165
You And I - SAINT WKND Remix 0.793 0.007 14.937
Flames - Extended 0.717 -0.009 96.702
Better When You’re Gone - Extended Mix 0.809 0.014 146.992
OctaHate - Cashmere Cat Remix 0.687 0.015 57.600
Focus - Yaeji Remix 0.614 0.016 -4.187
Don’t Start Now - Live in LA Remix 0.810 0.017 156.785
Toxic - Bloodshy & Avant’s Intoxicated Remix - 2009 Remaster 0.855 0.017 136.400

These remixes don’t feel that different from the original. Half of the songs on this list are mostly remixed that are just extended versions of the original song, namely Circus, Flames, Better When You’re Gone, Don’t Start Now, and Toxic. It makes sense that an extended version of a song is not drastically different from the original. Also interesting is that Mura Masa has a track in both the top 10 biggest and top 10 smallest changes in energy.

Structure of a remix through a chromagram.


Flume’s remix of Disclosure’s “You & Me” (featuring Eliza Doolittle) had the biggest change in energy in the corpus compared to its original recording. It’s also way more popular than the original. As of writing, the remix has 391 million Spotify streams, while the original has 19 million.

A chromagram shows how the pitch content is distributed between pitch classes (chroma’s) over time, this way harmonic and melodic characteristics can be captured and visualized. This one uses a Euclidean norm for the chroma vectors. Certain song structure elements are annotated on the x-axis.